PSCI 3300.003 Political Science Research Methods
A. Jordan Nafa
University of North Texas
September 1st, 2022
Political science is a field characterized by a diverse range of approaches to inquiry and debates about how we ought to study political phenomena have long animated the discipline.
Normative
Empirical
\[ \definecolor{treatment}{RGB}{255, 53, 94} \definecolor{treat}{RGB}{253, 91, 120} \definecolor{orange}{RGB}{255, 96, 55} \definecolor{confounders}{RGB}{255, 153, 51} \definecolor{lime}{RGB}{204, 255, 0} \definecolor{resp}{RGB}{102, 255, 102} \definecolor{index}{RGB}{170, 240, 209} \definecolor{untreat}{RGB}{80, 191, 230} \definecolor{pink}{RGB}{255, 110, 255} \definecolor{sample}{RGB}{255, 0, 204} \definecolor{operator}{RGB}{255,255,255} \]
Normative approaches to the study of politics date back thousands of years and feature prominently in the sub-field of political philosophy.
How should the world look? Asks for a moral judgement
Who should be responsible for paying for the consequences of climate change?
Should we fire Elon Musk into the Sun?
Should women have autonomy over their reproductive choices?
Is it fair to forgive student loan debt?
Normative arguments are common in certain areas of law and philosophy but have no place in this course as they do not lend themselves to scientific answers
Empirical approaches are those that aim to apply the scientific method to the study of politics and hold a dominant place contemporary political science.
Empirical approaches can be descriptive or causal, quantitative or qualitative, experimental or observational but they all aim to answer some question about how, what, or why the world is.
Description focuses on observing and measuring the state of the world; it aims to answer questions about who or what in relation to some phenomena (Gerring 2012).
What is democracy and how can we operationalize it?
Who won the 2020 presidential election election?
Description has a valuable place in political science and accurate description is essential to empirical research.
Descriptive approaches tend to lend themselves to dichotomous answers
Is country \(A\) more democratic than country \(B\)?
Is American democracy in decline?
Did Donald Trump lose the 2020 U.S. presidential election?
Yet, it is necessarily inferior to causal approaches because it cannot answer questions of why or how things happen
We’ll take a more detailed look at the role of description in quantitative political research when we discuss measurement
Causal approaches are concerned with explaining why some phenomenon occurs in the world (Samii 2016).
Contemporary political science is a discipline interested in answering causal questions.
Why do poor conservatives tend to vote against their own economic interests?
How do gender-inclusive peace processes influence the risk of conflict recurrence?
How would the world change if we fired Elon Musk into the Sun?
Our focus in this class will be primarily on causal questions and entirely on empirical approaches to the study of politics
Does forgiving student loan debt increase inflation?
Imagine student loan debt is forgiven and inflation increases
Would this increase have happened if student loan debt had not been forgiven?
How do gender-inclusive peace processes influence the risk of conflict recurrence?
Conflicts that terminate with gender-inclusive peace provisions tend to be less likely to recurr
Would conflict have recurred in the abscence of these gender-inclusive peace provisions?
Causal inference is about counterfactuals
A counterfactual is what would have happened in the absence of some intervention.
Imagine a study of \(\color{sample} n\) individuals
\(\color{sample} n_{\color{treat} 1}\) are assigned some treatment
\(\color{sample} n_{\color{untreat} 0}\) do not receive the treatment
For each individual \(\color{index} i \color{operator}\in \{1, 2, \dots, \color{sample} n \color{operator}\}\) we observe the outcome \(\color{resp}Y_{\color{index}i}\)
Treatment status for each individual \(\color{index} i\) \[\color{treatment} X_{\color{index}i} \color{operator} = \begin{cases}\color{treat} 1 \text{ if treated}\\ \color{untreat} 0 \text{ if not treated}\end{cases}\]
Some set of pre-treatment covariates \(\color{confounders} Z_{\color{index}i}\)
Counterfactuals are questions about the data we do not observe, not the data we do.
We want to know the causal effect of \(\color{treatment} X_{\color{index} i}\) on \(\color{resp} Y_{\color{index} i}\)
If an individual is treated, \(\color{treatment} X_{\color{index} i} \color{operator} = \color{treat} 1\) and we observe some value of \(\color{resp} Y_{\color{index} i}\)
What value of \(\color{resp} Y_{\color{index} i}\) would we have observed if \(\color{treatment} X_{\color{index} i} \color{operator} = \color{untreat} 0\) instead?
Fundamental Problem of Causal Inference
For each individual \(\color{index} i\) we can only observe \(\color{treatment}X_{\color{index} i} \color{operator} = \color{treat}1\) or \(\color{treatment}X_{\color{index} i} \color{operator} = \color{untreat} 0\)
Causal inference is a missing data problem
How do we overcome this problem?
Imagine we are interested in whether firing billionaires into the Sun might cause some meaningful improvement in the world.
A study of \(\color{sample} n\) billionaires
For each billionaire \(\color{index} i \color{operator}\in \{1, 2, \dots, \color{sample} n \color{operator}\}\) we observe the state of the world \(\color{resp}Y_{\color{index}i}\) before and after they are assigned to either the treatment or control group
Treatment status for each billionaire \(\color{index} i\) \[\color{treatment} X_{\color{index}i} \color{operator} = \begin{cases}\color{treat} 1 \text{ if fired into the Sun}\\ \color{untreat} 0 \text{ if not fired into the Sun}\end{cases}\]
Some set of pre-treatment covariates \(\color{confounders} Z_{\color{index}i}\)
The causal effect of firing a billionaire into the Sun is
For each billionaire \(\color{index} i\) we can either fire them into the Sun or not fire them into the Sun, but it is impossible do both
Also illustrates some practical limitations
No IRB would ever approve this study and billionaires are a hard to sample population 😭
More efficient to eject them out of the Solar system with a gravity assist from Jupiter instead
Descriptions of how the world is, correlations, joint distributions, predictions, regression coefficients, odds ratios, probabilities, etc.
All of these things may be useful and some may have causal interpretations under specific circumstances
They do not, however, in and of themselves capture causal relationships without additional assumptions
A causal effect is the change we would observe if we manipulated some feature of the world while holding all else constant
As it turns out, causal inference is really, really hard so why bother at all?
We could just make some claims and use a bunch of weasel words to avoid saying “cause” and “effect” while still heavily implying causality, right?
“\(\color{treatment}X\) explains \(\color{resp}Y\)”
“\(\color{treatment}X\) has an impact on \(\color{resp}Y\)”
“People who do \(\color{treatment}X\) are more likely to experience \(\color{resp}Y\)”
Lots of people still do this!
Makes it hard to distinguish between what is real and what is not, results in the proliferation of pseudo-facts (Samii 2016)
Important to be explicit about our assumptions, intentions, and goals
Difference between normative and empirical approaches in political science
Counterfactuals and the Fundamental Problem of Causal Inference
What causal inference is and what it is not
We will spend the rest of the semester building on this foundation
Read the installation instructions on Canvas under the module for Week I
Easiest way to avoid headaches involving file paths is for you to download the course’s repository from github
Either download a zip file of the current version and extract that somewhere on your computer
Or you can install git which integrates with RStudio
Happy Git and GitHub for the useR is a free online book on getting started with git in RStudio
This approach let’s you pull updates from the course repo directly from RStudio
This will ensure your relative file paths match mine and will make it easier for me to help you
Project-Oriented Workflow
Work in RStudio projects
Helps you keep things organized in appropriate subfolders
All file paths are relative to the .Rproj file’s location
Write code in scripts
Helps you keep track of and structure your data, analysis, etc.
Comment your code
Use seperate scripts for each part of your analysis, problem sets, etc.
We’ll talk more about Quarto and dynamically reproducible documents next week
In the simplest illustration, we can use R for both basic calculations like addition, subtraction, multiplication, and division
In the simplest illustration, we can use R for both basic calculations like addition, subtraction, multiplication, and division
In the simplest illustration, we can use R for both basic calculations like addition, subtraction, multiplication, and division
In the simplest illustration, we can use R for both basic calculations like addition, subtraction, multiplication, and division